Alternative Methods to Predict Fish Proximate Composition

We used a multiple linear regression approach to develop models predicting water, protein, and lipid content of bluegills (Lepomis macrochirus) under 4 measurement approaches varying in terms of time and money. Inputs were length, weight, relative weight, total body electrical conductivity, and water. Models predicting water and protein weights were very accurate (<5% mean error). No regression predicting lipid weight was accurate enough to be used as a predictor (>37% mean error). We then attempted to reduce inaccuracy by standardizing lipid weight 4 ways. No standardization substantially improved predictive accuracy (>30% mean error). However, our results suggest that increasing the range of values used to fit the regressions may increase precision and accuracy of prediction.

Starting page
110
Ending page
118
ID
12152